Yes, One-Bit-Flip Matters! Universal DNN Model Inference Depletion with Runtime Code Fault Injection
Bibliographic record
Abstract
We propose, FrameFlip, a novel attack for depleting DNN model inference with runtime code fault injections. Notably, Frameflip operates independently of the DNN models deployed and succeeds with only a single bit-flip injection. This fundamentally distinguishes it from the existing DNN inference depletion paradigm that requires injecting tens of deterministic faults concurrently. Since our attack performs at the universal code or library level, the mandatory code snippet can be perversely called by all mainstream machine learning frameworks, such as PyTorch and TensorFlow, dependent on the library code. Using DRAM Rowhammer to facilitate end-to-end fault injection, we implement Frameflip across diverse model architectures (LeNet, VGG-16, ResNet-34 and ResNet-50) with different datasets (FMNIST, CIFAR-10, GTSRB, and ImageNet). With a single bit fault injection, Frameflip achieves high depletion efficacy that consistently renders the model inference utility as no better than guessing. We also experimentally verify that identified vulnerable bits are almost equally effective at depleting different deployed models. In contrast, transferability is unattainable for all existing state-of-the-art model inference depletion attacks. Frameflip is shown to be evasive against all known defenses, generally due to the nature of current defenses operating at the model level (which is model-dependent) in lieu of the underlying code level.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.011 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.001 | 0.002 |
| Scholarly communication | 0.001 | 0.004 |
| Open science | 0.002 | 0.003 |
| Research integrity | 0.001 | 0.003 |
| Insufficient payload (model declined to judge) | 0.004 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".